Domain-Assisted Product Aspect Hierarchy Generation: Towards Hierarchical Organization of Unstructured Consumer Reviews

Proceedings of the 2011 Conference on Empirical Methods in Natural Language Processing, pages 140–150, Edinburgh, Scotland, UK, July 27–31, 2011. c⃝2011 Association for Computational Linguistics Domain-Assisted Product Aspect Hierarchy Generation: Towards Hierarchical Organization of Unstructured Consumer Reviews Jianxing Yu1, Zheng-Jun Zha1, Meng Wang1, Kai Wang2, Tat-Seng Chua1 1School of Computing, National University of Singapore 2Institute for Infocomm Research, Singapore {jianxing, zhazj, wangm, chuats}@comp.nus.edu.sg kwang@i2r.a-star.edu.sg Abstract This paper presents a domain-assisted ap- proach to organize various aspects of a prod- uct into a hierarchy by integrating domain knowledge (e.g., the product specifications), as well as consumer reviews. Based on the derived hierarchy, we generate a hierarchical organization of consumer reviews on various product aspects and aggregate consumer opin- ions on these aspects. With such organiza- tion, user can easily grasp the overview of consumer reviews. Furthermore, we apply the hierarchy to the task of implicit aspect identi- fication which aims to infer implicit aspects of the reviews that do not explicitly express those aspects but actually comment on them. The experimental results on 11 popular products in four domains demonstrate the effectiveness of our approach. 1 Introduction With the rapidly expanding e-commerce, most retail Web sites encourage consumers to write reviews to express their opinions on various aspects of prod- ucts. Huge collections of consumer reviews are now available on the Web. These reviews have be- come an important resource for both consumers and firms. Consumers commonly seek quality informa- tion from online consumer reviews prior to purchas- ing a product, while many firms use online reviews as an important resource in their product develop- ment, marketing, and consumer relationship man- agement. However, the reviews are disorganized, leading to the difficulty in information navigation and knowledge acquisition. It is impractical for user to grasp the overview of consumer reviews and opin- ions on various aspects of a product from such enor- mous reviews. Among hundreds of product aspects, it is also inefficient for user to browse consumer re- views and opinions on a specific aspect. Thus, there is a compelling need to organize consumer reviews, so as to transform the reviews into a useful knowl- edge structure. Since the hierarchy can improve in- formation representation and accessibility (Cimiano, 2006), we propose to organize the aspects of a prod- uct into a hierarchy and generate a hierarchical or- ganization of consumer reviews accordingly. Towards automatically deriving an aspect hierar- chy from the reviews, we could refer to traditional hierarchy generation methods in ontology learning, which first identify concepts from the text, then determine the parent-child relations between these concepts using either pattern-based or clustering- based methods (Murthy et al., 2010). However, pattern-based methods usually suffer from inconsis- tency of parent-child relationships among the con- cepts, while clustering-based methods often result in low accuracy. Thus, by directly utilizing these methods to generate an aspect hierarchy from con- sumer reviews, the resulting hierarchy is usually in- accurate, leading to unsatisfactory review organiza- tion. On the other hand, domain knowledge of prod- ucts is now …

Content truncated to 500 words